add fetch image api

.gitignore

@@ -0,0 +1 @@
+model_final.pth

app.py

@@ -97,7 +97,70 @@ def upload():
     class_ids = instances.pred_classes.numpy()
 
     # Filter overlapping boxes using IoU
-    iou_threshold = 0.5
+    iou_threshold = 0.8
+    keep_indices = []
+    merged_boxes = set()
+
+    for i in range(len(boxes)):
+        if i in merged_boxes:
+            continue
+        keep_indices.append(i)
+        for j in range(i + 1, len(boxes)):
+            if j in merged_boxes:
+                continue
+            iou = box_iou(
+                torch.tensor(boxes[i]).unsqueeze(0), torch.tensor(boxes[j]).unsqueeze(0)
+            ).item()
+            if iou > iou_threshold:
+                merged_boxes.add(j)
+
+    # Calculate total cost based on non-overlapping boxes
+    total_cost = 0
+    damage_details = []
+
+    for idx in keep_indices:
+        class_id = class_ids[idx]
+        damaged_part = (
+            class_names[class_id] if class_id < len(class_names) else "unknown"
+        )
+        if damaged_part not in expected_parts:
+            damaged_part = "other"
+
+        repair_cost = cost_dict.get(damaged_part, cost_dict["other"])
+        total_cost += repair_cost
+
+        damage_details.append({"part": damaged_part, "cost_usd": repair_cost})
+
+    response = {"damages": damage_details, "total_cost": total_cost}
+
+    return jsonify(response)
+
+
+@app.route("/fetch-image", methods=["POST"])
+def fetchImage():
+    file = None
+    if "url" in request.form:
+        url = request.form["url"]
+        response = requests.get(url)
+        file = io.BytesIO(response.content)
+    elif "file" in request.files:
+        file = request.files["file"]
+
+    # Load image
+    image = io.imread(file)
+    image_np = image
+
+    # Run model prediction
+    outputs = predictor(image_np)
+    instances = outputs["instances"].to("cpu")
+    class_names = MetadataCatalog.get(cfg.DATASETS.TEST[0]).thing_classes
+
+    # Extract bounding boxes and class IDs
+    boxes = instances.pred_boxes.tensor.numpy()
+    class_ids = instances.pred_classes.numpy()
+
+    # Filter overlapping boxes using IoU
+    iou_threshold = 0.8
     keep_indices = []
     merged_boxes = set()